Eight Prompts To Remove The Robot. None To Add The Person.
Eight prompts for making AI writing sound human, posted in the comments. Direct Rewrite. Anti-Robot Rewrite. Final Human-Touch Polish. Natural Voice Converter. Professional AI Humanizer. Emotion and Flow Editor. Natural Rhythm Editor. Simple Language Converter.
Read their verbs. Eliminate stiff language. Eliminate obvious AI patterns. Eliminate anything robotic, generic, repetitive. Avoid overly polished language. Replace generic phrasing. Remove flat or monotonous phrasing.
The word “eliminate” appears in six of the eight. Every single prompt is subtractive – it describes things to take out of a draft.
And every single one ends with the same line:
Text: [paste text]
That is the only input. The draft, and nothing else. Nowhere in any of the eight is there a slot for you – no question about what you know, who you work with, what you have seen go wrong, or what you believe that most people in your field do not.
Which matters, because that is the thing that actually makes writing sound like a person.
The post this article is about
Read the original post: “AI can write the words. But these prompts make them sound human.”
All eight prompts are posted in the comments beneath. The tables below quote them as written, read on 11 October 2026.
What actually makes writing sound like a person
Not the absence of robot-tells. Writing reads as human when it contains things that could only have come from one particular person – and a rewrite cannot invent those, because they are not in the draft to begin with.
Diagram: what a rewrite can and cannot add
| What makes it sound human | An example | Can a rewrite add it? | Which of the eight prompts touch it |
|---|---|---|---|
| A specific only you have | “We tried this on 40 thumbnails in August. Eleven of them did worse.” | No. It is not in the draft, and inventing one would be a lie | None of the eight |
| An opinion with a loser | “Resolve beats Premiere for this, and the free version is enough.” | No. A rewrite will soften an opinion, not supply one | None of the eight |
| Something you got wrong | “I assumed the numbers were inflated. Three of four were too low.” | No | None of the eight |
| A named thing instead of a category | “a crooked back tooth”, not “natural imperfections” | Partly. It can swap vague words for plainer ones, but it cannot know which tooth | 8 (Simple Language Converter) |
| Rhythm – varied sentence length | A nine-word sentence after a thirty-word one | Yes. Genuinely, and well | 2 and 7 – “adjust sentence lengths”, “improve the cadence” |
| Plain transitions | “But” and “so”, instead of “Moreover” and “Furthermore” | Yes | 2, 4, 6, 7 |
Rows one to three are the ones that do the work, and a rewrite cannot reach any of them. Rows five and six are real and the prompts handle them well – “adjust sentence lengths” is the single most useful instruction in the set, because uniform sentence length is a genuine, measurable tell and varying it genuinely helps.
But notice the shape of the result. Fix rhythm and transitions on a paragraph that contains no specifics, no opinion and no mistake, and you have produced a fluent paragraph that says the same nothing. It reads better. It is not more human. It is better-dressed.
All eight, with the add column filled in
Column 3 is full. Column 4 is almost entirely empty, and the one entry in it is the one to be careful with.
| Prompt | Quoted as written | What it removes | What it adds | Our note – read 11 October 2026 |
|---|---|---|---|---|
| 1. Direct Rewrite | “Rewrite this content as if a knowledgeable person were explaining it directly to another. Make it conversational, clear, warm, and engaging while remaining professional. Eliminate stiff, unnatural, or overly formal language.” | Stiffness, formality | Nothing | “As if a knowledgeable person” – as if. The prompt asks for the register of expertise without any of the expertise. That gap is the whole post |
| 2. Anti-Robot Rewrite | “Eliminate obvious AI writing patterns. Adjust sentence structure and length, use natural transitions, and replace generic or repetitive phrasing with authentic language.” | AI patterns, uniform sentences | Varied rhythm, transitions | The most technically useful of the eight. “Adjust sentence structure and length” is a concrete, checkable instruction rather than an adjective |
| 3. Final Human-Touch Polish | “Refine the tone, structure, flow, word choice, and sentence rhythm. Eliminate anything robotic, generic, repetitive, or overly formal, while retaining the original meaning and personality.” | Everything the others remove | Nothing new | A superset of 1, 4, 5 and 7. Note “retaining the original personality” – it can only retain a personality the draft already had |
| 4. Natural Voice Converter | “Rewrite this text in a natural, confident voice—as if written by someone with genuine experience. Improve word choice, sentence rhythm, and transitions.” | Over-polish | Confidence | “As if written by someone with genuine experience.” Confidence without experience is the most recognisable failure mode in AI prose, and this prompt asks for exactly that |
| 5. Professional AI Humanizer | “Act as an expert editor… Eliminate robotic wording, predictable patterns, and awkward phrasing while preserving the original meaning, facts, and intent.” | Robotic wording | Nothing | The best guard clause in the set: “preserving the original meaning, facts, and intent”. It is the only one that names facts, which is what stops a rewrite quietly changing a number |
| 6. Emotion and Flow Editor | “Edit this text to infuse it with personality, emotional depth, and natural flow… Keep the emotion authentic and avoid overdoing it.” | Flatness | Personality and emotion | The only additive prompt, and the riskiest. “Infuse” emotion into a text with no story in it and you get emphasis without substance. Authentic to what? There is no source. The prompt half-knows: “avoid overdoing it” |
| 7. Natural Rhythm Editor | “Adjust sentence lengths, improve the cadence, eliminate repetitive structures, and ensure each sentence flows naturally into the next. Preserve the author’s original voice and message.” | Monotone cadence | Rhythm | The most honest prompt here. It claims one narrow thing, does it well, and says “preserve the author’s original voice” rather than pretending to create one |
| 8. Simple Language Converter | “Replace complex vocabulary, convoluted phrasing, and overly formal expressions with words real people actually use. Keep the meaning, accuracy, and key details intact.” | Jargon, complexity | Plain words | Underrated. Plain vocabulary does more for readability than any of the “sound human” instructions, and “keep the accuracy and key details intact” is a proper guard |
Two things called humanizing, and only one is editing
Worth naming the line plainly, because the word covers two activities.
One is editing. You wrote something, or you drafted it with help, and the prose is stiff. You cut the jargon, vary the sentence lengths, fix the transitions. That is what every editor who ever lived has done, and these eight prompts are a decent editor. The card frames it this way throughout – “preserving the original meaning, facts, and intent”, “preserve the author’s original voice and message”.
The other is evasion. Making machine-written text pass as human-written somewhere that matters: coursework, a client who is paying for a person, a platform whose policy says otherwise. These prompts would serve that too, because the mechanics are identical – the only difference is whether the thinking underneath is yours.
The test is simple and it is not about the words. If you could defend every claim in the finished piece without reading it again, it is edited. If you could not, it is laundered.
For anyone publishing video, there is a concrete version of this: we have twice read YouTube’s page on disclosing GenAI content while writing this series, and its four triggers are about realistic synthetic depictions, not about whether an editor tightened your script. Polishing your own words is not a disclosure event. Passing off a machine’s opinions as your experience is a different problem, and no platform policy is what makes it one.
The prompt the set is missing
If seven of the eight remove things and one adds invented emotion, the gap is an additive prompt that adds true things. It cannot work by rewriting, because the material is not in the draft. It has to interview you.
| The question it should ask you | Why this one | What it puts into the text | Roughly how long |
|---|---|---|---|
| 1. “What number do you remember?” | A real figure cannot be generated. It can only be recalled | The one sentence nobody else on the internet could have written | 10 seconds |
| 2. “Who said this to you, and in what words?” | A quoted client or viewer is unfakeable specificity | A second voice in the piece, which breaks the monotone better than any rhythm edit | 20 seconds |
| 3. “What did you believe before, and what changed it?” | A reversal is the strongest human signal there is – machines do not change their minds in public | A before and an after, which is a story shape | 30 seconds |
| 4. “What is the common advice here that you think is wrong?” | An opinion with a loser. Prompt 6 asks for “personality”; this is where personality actually comes from | A position somebody could disagree with | 30 seconds |
| 5. “What went wrong when you tried it?” | Failures are specific, dated and nobody fabricates them in their own favour | Credibility, which cannot be added by word choice | 30 seconds |
| Then, and only then: run prompt 2 or 7 | Rhythm editing works beautifully on text that has something in it | A piece that reads well and says something | One pass |
Two minutes of honest answers put more humanity into a draft than all eight prompts combined, because they add material rather than rearranging it. And they make the eight work better – a rhythm edit on a paragraph containing a real number and a real mistake produces something genuinely good, which is exactly what prompt 7 is for.
The order matters. Add first, subtract second. Running the subtractive prompts on an empty draft just polishes the emptiness, which is the same error we found in a card of ten video tools where nothing wrote the script.
Four honest caveats
These are good editing prompts and we would use two of them. Prompt 2 and prompt 7 both say “adjust sentence lengths” – a concrete, checkable instruction rather than an adjective, and uniform sentence length is a real tell that genuinely improves when you fix it. Prompt 8 is underrated: plain vocabulary does more for readability than anything else in the set. Our argument is that the collection is incomplete, not that it is bad.
Prompt 5 has the best guard clause in the set. “Preserving the original meaning, facts, and intent.” It is the only one of the eight that names facts specifically, and a rewrite that quietly rounds a number or softens a claim is a real risk when you are running eight passes over the same paragraph. Add that clause to the other seven.
Eight is more than one. Prompt 3 is close to a superset of 1, 4, 5 and 7 – every one of those asks for tone, flow, word choice and rhythm with different adjectives attached. If you only keep two, keep 7 (rhythm, honestly scoped) and 8 (plain words). Running all eight in sequence on one paragraph will sand it down to something smooth and characterless, which is the opposite of the stated goal.
And we have not run them. Everything here is read off the prompts as written. We have not measured what any particular model does with them, and results will differ by model and by draft.
The one-line version
Eight prompts, six uses of the word “eliminate”, and one input line repeated eight times: Text: [paste text]. Every one of them works on the draft you already have, and the draft is exactly where the problem is not.
Writing sounds human when it carries things only one person could supply – a number you remember, a sentence somebody actually said to you, a thing you used to believe and no longer do, an opinion with a loser, a failure you are willing to describe. No rewrite can add any of those, because none of them is in the text to begin with.
So invert the order. Spend two minutes answering five questions about what you actually know, put the answers in, and then run prompt 7 for rhythm and prompt 8 for plain words. Add first, subtract second.
And keep the test in your pocket: if you could defend every claim in the finished piece without reading it again, you edited it. If you could not, you only dressed it.
Learning a lot does not make anyone good. Doing a lot does. Running eight humanizer prompts on an empty paragraph is eight ways to be busy. Writing down the one number you actually remember takes ten seconds and does more than all of them.
If you want to be held to that, we run a free challenge at BANI Academy. There is no barrier to entry – you commit a small fee that is refunded when you finish, and the only requirement is that you actually do the work for the length of the challenge.
Take something you published this month and answer question three: what did you believe before, and what changed it? Put the answer in the comments under the original post – that sentence is probably better than the paragraph it would replace.
More on building a channel for a global audience at mmoyoutube.com.
2 Comments